From ed09cac44052517d2cf08fe1ebc00c6a5b51e546 Mon Sep 17 00:00:00 2001 From: Triang3l Date: Fri, 28 Dec 2018 08:11:34 +0300 Subject: [PATCH] [D3D12] DXBC: vfetch RoundIndex, vertex index signedness fixes --- .../gpu/d3d12/d3d12_command_processor.cc | 2 +- .../d3d12/shaders/dxbc/continuous_quad_hs.cso | Bin 4236 -> 4236 bytes .../d3d12/shaders/dxbc/continuous_quad_hs.h | 56 +++++++++--------- .../d3d12/shaders/dxbc/continuous_quad_hs.txt | 2 +- .../shaders/dxbc/continuous_triangle_hs.cso | Bin 4092 -> 4092 bytes .../shaders/dxbc/continuous_triangle_hs.h | 56 +++++++++--------- .../shaders/dxbc/continuous_triangle_hs.txt | 2 +- .../d3d12/shaders/dxbc/discrete_quad_hs.cso | Bin 4236 -> 4236 bytes .../gpu/d3d12/shaders/dxbc/discrete_quad_hs.h | 56 +++++++++--------- .../d3d12/shaders/dxbc/discrete_quad_hs.txt | 2 +- .../shaders/dxbc/discrete_triangle_hs.cso | Bin 4092 -> 4092 bytes .../d3d12/shaders/dxbc/discrete_triangle_hs.h | 56 +++++++++--------- .../shaders/dxbc/discrete_triangle_hs.txt | 2 +- .../shaders/dxbc/primitive_point_list_gs.cso | Bin 7792 -> 7792 bytes .../shaders/dxbc/primitive_point_list_gs.h | 54 ++++++++--------- .../shaders/dxbc/primitive_point_list_gs.txt | 2 +- src/xenia/gpu/d3d12/shaders/xenos_draw.hlsli | 2 +- src/xenia/gpu/dxbc_shader_translator.cc | 4 +- src/xenia/gpu/dxbc_shader_translator.h | 2 +- src/xenia/gpu/dxbc_shader_translator_fetch.cc | 54 +++++++++++++---- 20 files changed, 191 insertions(+), 161 deletions(-) diff --git a/src/xenia/gpu/d3d12/d3d12_command_processor.cc b/src/xenia/gpu/d3d12/d3d12_command_processor.cc index d492d849a..9d474b90e 100644 --- a/src/xenia/gpu/d3d12/d3d12_command_processor.cc +++ b/src/xenia/gpu/d3d12/d3d12_command_processor.cc @@ -1905,7 +1905,7 @@ void D3D12CommandProcessor::UpdateSystemConstantValues( SCOPE_profile_cpu_f("gpu"); #endif // FINE_GRAINED_DRAW_SCOPES - uint32_t vgt_indx_offset = regs[XE_GPU_REG_VGT_INDX_OFFSET].u32; + int32_t vgt_indx_offset = int32_t(regs[XE_GPU_REG_VGT_INDX_OFFSET].u32); uint32_t pa_cl_vte_cntl = regs[XE_GPU_REG_PA_CL_VTE_CNTL].u32; uint32_t rb_depthcontrol = regs[XE_GPU_REG_RB_DEPTHCONTROL].u32; uint32_t rb_stencilrefmask = regs[XE_GPU_REG_RB_STENCILREFMASK].u32; diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.cso b/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.cso index 1744606e09e8f73204da29d1d35dc199be9ea4fb..0d83b4215ef8068d097355fe55fea69db8b40972 100644 GIT binary patch delta 166 zcmeBC>`@ePiEwfbm~~xXk%IJc*%{9RPuwzf+bAN#$fz>8kx`u0gM)#AXX0OF!AKxO zfC0z>Vh|_*NefN^ipl_aEMPGPh8`@ePiEwg`HfPBaojSYzg!JCO?t70`ZWNJWoLs>o$-=?Fz%%i$vY;W5A;19S z05J%7Og3Z`7t8^Q$^dyRU@-=U31HDJKv4yt2&&pcKv5kewO2rDHw!QYuubmZ@L;r? be38SQv3;^6r~Kpu4i)AL91NSyIYZe2=zSlS diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.h b/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.h index 034e15d58..06b02f512 100644 --- a/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.h +++ b/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.h @@ -1,8 +1,8 @@ // generated from `xb buildhlsl` // source: continuous_quad.hs.hlsl const uint8_t continuous_quad_hs[] = { - 0x44, 0x58, 0x42, 0x43, 0x5B, 0x37, 0x04, 0x6A, 0x15, 0x95, 0x9B, 0x7F, - 0xC8, 0x1B, 0xBD, 0xFD, 0x47, 0xBD, 0xC5, 0x79, 0x01, 0x00, 0x00, 0x00, + 0x44, 0x58, 0x42, 0x43, 0x50, 0x9A, 0xD7, 0x10, 0xA2, 0x20, 0x1B, 0xA7, + 0x1D, 0x98, 0xE7, 0x51, 0xC8, 0xDA, 0x35, 0x46, 0x01, 0x00, 0x00, 0x00, 0x8C, 0x10, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0xA0, 0x0D, 0x00, 0x00, 0xB0, 0x0D, 0x00, 0x00, 0xC0, 0x0D, 0x00, 0x00, 0x84, 0x0E, 0x00, 0x00, 0xF0, 0x0F, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, @@ -26,24 +26,24 @@ const uint8_t continuous_quad_hs[] = { 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x0B, 0x08, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x24, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, - 0x00, 0x00, 0x00, 0x00, 0x20, 0x08, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x48, 0x08, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x31, 0x08, 0x00, 0x00, + 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x59, 0x08, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x48, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, + 0x70, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0x6C, 0x08, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x90, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x94, 0x08, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0xB8, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, - 0x00, 0x00, 0x00, 0x00, 0xB4, 0x08, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, - 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x08, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0xDC, 0x08, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, + 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x70, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xC2, 0x08, 0x00, 0x00, + 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xEA, 0x08, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0xD4, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, + 0x24, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xF8, 0x08, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x10, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, @@ -179,25 +179,25 @@ const uint8_t continuous_quad_hs[] = { 0x78, 0x65, 0x5F, 0x76, 0x65, 0x72, 0x74, 0x65, 0x78, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x5F, 0x65, 0x6E, 0x64, 0x69, 0x61, 0x6E, 0x00, 0x78, 0x65, 0x5F, 0x76, 0x65, 0x72, 0x74, 0x65, 0x78, 0x5F, 0x62, 0x61, 0x73, - 0x65, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x00, 0x78, 0x65, 0x5F, 0x70, - 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x70, 0x6F, 0x73, 0x5F, 0x72, 0x65, 0x67, - 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x73, 0x63, 0x61, 0x6C, - 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x33, 0x00, 0xAB, 0xAB, 0xAB, - 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x65, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x00, 0x69, 0x6E, 0x74, 0x00, + 0x00, 0x00, 0x02, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3E, 0x08, 0x00, 0x00, - 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x68, 0x61, 0x6C, - 0x66, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x6F, 0x66, 0x66, 0x73, - 0x65, 0x74, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x00, 0xAB, 0xAB, 0xAB, - 0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x08, 0x00, 0x00, + 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x70, 0x6F, 0x73, + 0x5F, 0x72, 0x65, 0x67, 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, + 0x73, 0x63, 0x61, 0x6C, 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x33, + 0x00, 0xAB, 0xAB, 0xAB, 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x87, 0x08, 0x00, 0x00, - 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x6F, 0x66, 0x66, 0x73, 0x65, - 0x74, 0x00, 0x78, 0x65, 0x5F, 0x61, 0x6C, 0x70, 0x68, 0x61, 0x5F, 0x74, - 0x65, 0x73, 0x74, 0x00, 0x69, 0x6E, 0x74, 0x00, 0x00, 0x00, 0x02, 0x00, - 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0xD0, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, + 0x66, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, + 0x5F, 0x68, 0x61, 0x6C, 0x66, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, + 0x6F, 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, + 0x00, 0xAB, 0xAB, 0xAB, 0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0xAF, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x6F, + 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x78, 0x65, 0x5F, 0x61, 0x6C, 0x70, + 0x68, 0x61, 0x5F, 0x74, 0x65, 0x73, 0x74, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x6F, 0x69, 0x6E, 0x74, 0x5F, 0x73, 0x69, 0x7A, 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x32, 0x00, 0xAB, 0xAB, 0xAB, 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.txt b/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.txt index dc48f6e9f..b363a8151 100644 --- a/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.txt +++ b/src/xenia/gpu/d3d12/shaders/dxbc/continuous_quad_hs.txt @@ -9,7 +9,7 @@ // // uint xe_flags; // Offset: 0 Size: 4 [unused] // uint xe_vertex_index_endian; // Offset: 4 Size: 4 [unused] -// uint xe_vertex_base_index; // Offset: 8 Size: 4 [unused] +// int xe_vertex_base_index; // Offset: 8 Size: 4 [unused] // uint xe_pixel_pos_reg; // Offset: 12 Size: 4 [unused] // float3 xe_ndc_scale; // Offset: 16 Size: 12 [unused] // float xe_pixel_half_pixel_offset; // Offset: 28 Size: 4 [unused] diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/continuous_triangle_hs.cso b/src/xenia/gpu/d3d12/shaders/dxbc/continuous_triangle_hs.cso index ac01b47cce10d61781f7e749f3f5905108faed32..ddfa4ee8006e30036aa3b3de7249cbde4baad6e2 100644 GIT binary patch delta 166 zcmew(|3_ZLCBn)1OR=inZBx_r#}7|oTCM$L+eQ%?Mn;v%jf~=~9vln|JQM#a3q}GN z0t`S75Q9JgNLp|TP*euUV*!gXFzf(}-T{g#07X#Mz5rwX$I c2gBre4o}9k$tyYB8P`w#$RWSkg7XL)0Jv=bDg*>1;u?pk|vHj2nFPOe~)WZ__7;F>uuowfw1hD89pr`^+1Xb-Jpr{U#+AAQnn+2Ex*d}*ycre;c bzR2Ou*gjd3Q+{#+hYIrr4u;L`@ePiEwg`Hs{&(J7#@fMg5bhvWX=|8%1Oo8C51XGK#Z$a4<0NO#G`X7ztzu zFaS9~3<3opX~8K#Q5hhQ1uVwEumdc52Pmol6hT${3Mi_Bq!w(!W&x%Ew#fpVD$EKT b43pzIJQ>p_ujFuNTtE3Ehx}#>&JcD0z3d@n delta 163 zcmeBC>`@ePiEwgmycAW`Cv7!lcS_P+yR}h58%1OoCs(jYvT!gk@J#%xENBR12rvLS zKnwyNlMNZg1#^I+GC&>+Sd4*T0$6klP*edZf~xiqP*ev=?G=#P%>qmTY?C`UJQ(dJ bU*vFSY@aO2DL*-ZLxuSQ2g7D_&QNv$po1R% diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_quad_hs.h b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_quad_hs.h index 01c2b3aaf..8fcc3f0ad 100644 --- a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_quad_hs.h +++ b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_quad_hs.h @@ -1,8 +1,8 @@ // generated from `xb buildhlsl` // source: discrete_quad.hs.hlsl const uint8_t discrete_quad_hs[] = { - 0x44, 0x58, 0x42, 0x43, 0x81, 0xD2, 0x5A, 0x72, 0x8E, 0x1B, 0x3A, 0x94, - 0xBB, 0x64, 0x62, 0x9D, 0x3E, 0xAD, 0x5A, 0x12, 0x01, 0x00, 0x00, 0x00, + 0x44, 0x58, 0x42, 0x43, 0x5B, 0x37, 0x0C, 0xBA, 0xFB, 0x5C, 0xAF, 0x8E, + 0x78, 0x7F, 0xE4, 0x95, 0x1D, 0x61, 0x74, 0x32, 0x01, 0x00, 0x00, 0x00, 0x8C, 0x10, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0xA0, 0x0D, 0x00, 0x00, 0xB0, 0x0D, 0x00, 0x00, 0xC0, 0x0D, 0x00, 0x00, 0x84, 0x0E, 0x00, 0x00, 0xF0, 0x0F, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, @@ -26,24 +26,24 @@ const uint8_t discrete_quad_hs[] = { 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x0B, 0x08, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x24, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, - 0x00, 0x00, 0x00, 0x00, 0x20, 0x08, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x48, 0x08, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x31, 0x08, 0x00, 0x00, + 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x59, 0x08, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x48, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, + 0x70, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0x6C, 0x08, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x90, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x94, 0x08, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0xB8, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, - 0x00, 0x00, 0x00, 0x00, 0xB4, 0x08, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, - 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x08, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0xDC, 0x08, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, + 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x70, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xC2, 0x08, 0x00, 0x00, + 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xEA, 0x08, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0xD4, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, + 0x24, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xF8, 0x08, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x10, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, @@ -179,25 +179,25 @@ const uint8_t discrete_quad_hs[] = { 0x78, 0x65, 0x5F, 0x76, 0x65, 0x72, 0x74, 0x65, 0x78, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x5F, 0x65, 0x6E, 0x64, 0x69, 0x61, 0x6E, 0x00, 0x78, 0x65, 0x5F, 0x76, 0x65, 0x72, 0x74, 0x65, 0x78, 0x5F, 0x62, 0x61, 0x73, - 0x65, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x00, 0x78, 0x65, 0x5F, 0x70, - 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x70, 0x6F, 0x73, 0x5F, 0x72, 0x65, 0x67, - 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x73, 0x63, 0x61, 0x6C, - 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x33, 0x00, 0xAB, 0xAB, 0xAB, - 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x65, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x00, 0x69, 0x6E, 0x74, 0x00, + 0x00, 0x00, 0x02, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3E, 0x08, 0x00, 0x00, - 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x68, 0x61, 0x6C, - 0x66, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x6F, 0x66, 0x66, 0x73, - 0x65, 0x74, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x00, 0xAB, 0xAB, 0xAB, - 0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x08, 0x00, 0x00, + 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x70, 0x6F, 0x73, + 0x5F, 0x72, 0x65, 0x67, 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, + 0x73, 0x63, 0x61, 0x6C, 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x33, + 0x00, 0xAB, 0xAB, 0xAB, 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x87, 0x08, 0x00, 0x00, - 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x6F, 0x66, 0x66, 0x73, 0x65, - 0x74, 0x00, 0x78, 0x65, 0x5F, 0x61, 0x6C, 0x70, 0x68, 0x61, 0x5F, 0x74, - 0x65, 0x73, 0x74, 0x00, 0x69, 0x6E, 0x74, 0x00, 0x00, 0x00, 0x02, 0x00, - 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0xD0, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, + 0x66, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, + 0x5F, 0x68, 0x61, 0x6C, 0x66, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, + 0x6F, 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, + 0x00, 0xAB, 0xAB, 0xAB, 0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0xAF, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x6F, + 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x78, 0x65, 0x5F, 0x61, 0x6C, 0x70, + 0x68, 0x61, 0x5F, 0x74, 0x65, 0x73, 0x74, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x6F, 0x69, 0x6E, 0x74, 0x5F, 0x73, 0x69, 0x7A, 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x32, 0x00, 0xAB, 0xAB, 0xAB, 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_quad_hs.txt b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_quad_hs.txt index eae408426..6d9147940 100644 --- a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_quad_hs.txt +++ b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_quad_hs.txt @@ -9,7 +9,7 @@ // // uint xe_flags; // Offset: 0 Size: 4 [unused] // uint xe_vertex_index_endian; // Offset: 4 Size: 4 [unused] -// uint xe_vertex_base_index; // Offset: 8 Size: 4 [unused] +// int xe_vertex_base_index; // Offset: 8 Size: 4 [unused] // uint xe_pixel_pos_reg; // Offset: 12 Size: 4 [unused] // float3 xe_ndc_scale; // Offset: 16 Size: 12 [unused] // float xe_pixel_half_pixel_offset; // Offset: 28 Size: 4 [unused] diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.cso b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.cso index cd50f01bcd73e279bfd9a969cfd78f12b02ca51d..11c05145207485b9791d7a7fabc76949ff15d949 100644 GIT binary patch delta 166 zcmew(|3_ZLCBn(M+>uuowfw1hD89pr`^+1Xb-Jpr{U#+AAQnn+2Ex*d}*ycre;c bzR2Ou*gjd3Q+{#+hYIrr4u;Lc diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.h b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.h index 8430a972e..7f7f4b46a 100644 --- a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.h +++ b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.h @@ -1,8 +1,8 @@ // generated from `xb buildhlsl` // source: discrete_triangle.hs.hlsl const uint8_t discrete_triangle_hs[] = { - 0x44, 0x58, 0x42, 0x43, 0x1E, 0x53, 0x2C, 0x4C, 0xC6, 0xA1, 0x35, 0x71, - 0x0C, 0x7F, 0xBA, 0xC6, 0x8C, 0x86, 0x55, 0x76, 0x01, 0x00, 0x00, 0x00, + 0x44, 0x58, 0x42, 0x43, 0x77, 0x3F, 0xB5, 0xAC, 0x07, 0x09, 0x69, 0xF6, + 0x6D, 0x52, 0x1B, 0x6A, 0x91, 0x0F, 0xE9, 0x47, 0x01, 0x00, 0x00, 0x00, 0xFC, 0x0F, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0xA0, 0x0D, 0x00, 0x00, 0xB0, 0x0D, 0x00, 0x00, 0xC0, 0x0D, 0x00, 0x00, 0x54, 0x0E, 0x00, 0x00, 0x60, 0x0F, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, @@ -26,24 +26,24 @@ const uint8_t discrete_triangle_hs[] = { 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x0B, 0x08, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x24, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, - 0x00, 0x00, 0x00, 0x00, 0x20, 0x08, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x48, 0x08, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x31, 0x08, 0x00, 0x00, + 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x59, 0x08, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x48, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, + 0x70, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0x6C, 0x08, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x90, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x94, 0x08, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0xB8, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, - 0x00, 0x00, 0x00, 0x00, 0xB4, 0x08, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, - 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x08, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0xDC, 0x08, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, + 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x70, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xC2, 0x08, 0x00, 0x00, + 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xEA, 0x08, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0xD4, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, + 0x24, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xF8, 0x08, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x10, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, @@ -179,25 +179,25 @@ const uint8_t discrete_triangle_hs[] = { 0x78, 0x65, 0x5F, 0x76, 0x65, 0x72, 0x74, 0x65, 0x78, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x5F, 0x65, 0x6E, 0x64, 0x69, 0x61, 0x6E, 0x00, 0x78, 0x65, 0x5F, 0x76, 0x65, 0x72, 0x74, 0x65, 0x78, 0x5F, 0x62, 0x61, 0x73, - 0x65, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x00, 0x78, 0x65, 0x5F, 0x70, - 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x70, 0x6F, 0x73, 0x5F, 0x72, 0x65, 0x67, - 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x73, 0x63, 0x61, 0x6C, - 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x33, 0x00, 0xAB, 0xAB, 0xAB, - 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x65, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x00, 0x69, 0x6E, 0x74, 0x00, + 0x00, 0x00, 0x02, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3E, 0x08, 0x00, 0x00, - 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x68, 0x61, 0x6C, - 0x66, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x6F, 0x66, 0x66, 0x73, - 0x65, 0x74, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x00, 0xAB, 0xAB, 0xAB, - 0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x08, 0x00, 0x00, + 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x70, 0x6F, 0x73, + 0x5F, 0x72, 0x65, 0x67, 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, + 0x73, 0x63, 0x61, 0x6C, 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x33, + 0x00, 0xAB, 0xAB, 0xAB, 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x87, 0x08, 0x00, 0x00, - 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x6F, 0x66, 0x66, 0x73, 0x65, - 0x74, 0x00, 0x78, 0x65, 0x5F, 0x61, 0x6C, 0x70, 0x68, 0x61, 0x5F, 0x74, - 0x65, 0x73, 0x74, 0x00, 0x69, 0x6E, 0x74, 0x00, 0x00, 0x00, 0x02, 0x00, - 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0xD0, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, + 0x66, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, + 0x5F, 0x68, 0x61, 0x6C, 0x66, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, + 0x6F, 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, + 0x00, 0xAB, 0xAB, 0xAB, 0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0xAF, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x6F, + 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x78, 0x65, 0x5F, 0x61, 0x6C, 0x70, + 0x68, 0x61, 0x5F, 0x74, 0x65, 0x73, 0x74, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x6F, 0x69, 0x6E, 0x74, 0x5F, 0x73, 0x69, 0x7A, 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x32, 0x00, 0xAB, 0xAB, 0xAB, 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.txt b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.txt index 1762a970f..ad5028c81 100644 --- a/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.txt +++ b/src/xenia/gpu/d3d12/shaders/dxbc/discrete_triangle_hs.txt @@ -9,7 +9,7 @@ // // uint xe_flags; // Offset: 0 Size: 4 [unused] // uint xe_vertex_index_endian; // Offset: 4 Size: 4 [unused] -// uint xe_vertex_base_index; // Offset: 8 Size: 4 [unused] +// int xe_vertex_base_index; // Offset: 8 Size: 4 [unused] // uint xe_pixel_pos_reg; // Offset: 12 Size: 4 [unused] // float3 xe_ndc_scale; // Offset: 16 Size: 12 [unused] // float xe_pixel_half_pixel_offset; // Offset: 28 Size: 4 [unused] diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.cso b/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.cso index fed568f5c725fcf1ea88562eb6e43817ec085c01..7a1d3d9ec127dd681511cac7223bafc6e89045ec 100644 GIT binary patch delta 166 zcmexh^T9^MCBn)14Cg}Y#SAPOIpjB+aN39h0H{YH6951J delta 163 zcmexh^T9^MCBn(M>rC<=o_(_72hTkTzRRg7yir7gapD~bRs{|Q2A+u*m4ys}3;_lp z2Z%wygJa@7F`*ows0@(D0v2Onn7}c4ACtHsP*edZf~xiqP*ev=?G=y-n|YW5*e17d fcre;cKFi_G*gjd5Q+{#`hYIrr4u;LfoVKC>IE^Bx diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.h b/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.h index 9a8a33345..e8e970fa3 100644 --- a/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.h +++ b/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.h @@ -1,8 +1,8 @@ // generated from `xb buildhlsl` // source: primitive_point_list.gs.hlsl const uint8_t primitive_point_list_gs[] = { - 0x44, 0x58, 0x42, 0x43, 0x8A, 0xCC, 0x63, 0xFC, 0x0C, 0xBE, 0x1D, 0x17, - 0xC1, 0xCE, 0xE4, 0x53, 0xDD, 0x09, 0x21, 0x13, 0x01, 0x00, 0x00, 0x00, + 0x44, 0x58, 0x42, 0x43, 0xCC, 0x09, 0xA1, 0x3B, 0xA3, 0x00, 0x04, 0xA4, + 0x59, 0x7C, 0x61, 0x42, 0x24, 0x04, 0xBC, 0x91, 0x01, 0x00, 0x00, 0x00, 0x70, 0x1E, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00, 0x9C, 0x0D, 0x00, 0x00, 0x8C, 0x0F, 0x00, 0x00, 0xC8, 0x11, 0x00, 0x00, 0xD4, 0x1D, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0x60, 0x0D, 0x00, 0x00, @@ -26,23 +26,23 @@ const uint8_t primitive_point_list_gs[] = { 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x0B, 0x08, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, + 0x24, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0x20, 0x08, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, + 0x48, 0x08, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xD0, 0x07, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, - 0x00, 0x00, 0x00, 0x00, 0x31, 0x08, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, - 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x08, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x59, 0x08, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, + 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x70, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x6C, 0x08, 0x00, 0x00, + 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x94, 0x08, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x90, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, + 0xB8, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, - 0xB4, 0x08, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x48, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0xDC, 0x08, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x70, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, - 0x00, 0x00, 0x00, 0x00, 0xC2, 0x08, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, - 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xD4, 0x08, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0xEA, 0x08, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, + 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x24, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0xF8, 0x08, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, @@ -179,25 +179,25 @@ const uint8_t primitive_point_list_gs[] = { 0x65, 0x72, 0x74, 0x65, 0x78, 0x5F, 0x69, 0x6E, 0x64, 0x65, 0x78, 0x5F, 0x65, 0x6E, 0x64, 0x69, 0x61, 0x6E, 0x00, 0x78, 0x65, 0x5F, 0x76, 0x65, 0x72, 0x74, 0x65, 0x78, 0x5F, 0x62, 0x61, 0x73, 0x65, 0x5F, 0x69, 0x6E, - 0x64, 0x65, 0x78, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, - 0x5F, 0x70, 0x6F, 0x73, 0x5F, 0x72, 0x65, 0x67, 0x00, 0x78, 0x65, 0x5F, - 0x6E, 0x64, 0x63, 0x5F, 0x73, 0x63, 0x61, 0x6C, 0x65, 0x00, 0x66, 0x6C, - 0x6F, 0x61, 0x74, 0x33, 0x00, 0xAB, 0xAB, 0xAB, 0x01, 0x00, 0x03, 0x00, - 0x01, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x3E, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, - 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x68, 0x61, 0x6C, 0x66, 0x5F, 0x70, 0x69, - 0x78, 0x65, 0x6C, 0x5F, 0x6F, 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x66, - 0x6C, 0x6F, 0x61, 0x74, 0x00, 0xAB, 0xAB, 0xAB, 0x00, 0x00, 0x03, 0x00, + 0x64, 0x65, 0x78, 0x00, 0x69, 0x6E, 0x74, 0x00, 0x00, 0x00, 0x02, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0x00, 0x00, 0x00, 0x00, 0x87, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x6E, - 0x64, 0x63, 0x5F, 0x6F, 0x66, 0x66, 0x73, 0x65, 0x74, 0x00, 0x78, 0x65, - 0x5F, 0x61, 0x6C, 0x70, 0x68, 0x61, 0x5F, 0x74, 0x65, 0x73, 0x74, 0x00, - 0x69, 0x6E, 0x74, 0x00, 0x00, 0x00, 0x02, 0x00, 0x01, 0x00, 0x01, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x20, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, + 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x70, 0x6F, 0x73, 0x5F, 0x72, 0x65, 0x67, + 0x00, 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x73, 0x63, 0x61, 0x6C, + 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x33, 0x00, 0xAB, 0xAB, 0xAB, + 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x66, 0x08, 0x00, 0x00, + 0x78, 0x65, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x68, 0x61, 0x6C, + 0x66, 0x5F, 0x70, 0x69, 0x78, 0x65, 0x6C, 0x5F, 0x6F, 0x66, 0x66, 0x73, + 0x65, 0x74, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x00, 0xAB, 0xAB, 0xAB, + 0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, - 0xD0, 0x08, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x6F, 0x69, 0x6E, 0x74, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xAF, 0x08, 0x00, 0x00, + 0x78, 0x65, 0x5F, 0x6E, 0x64, 0x63, 0x5F, 0x6F, 0x66, 0x66, 0x73, 0x65, + 0x74, 0x00, 0x78, 0x65, 0x5F, 0x61, 0x6C, 0x70, 0x68, 0x61, 0x5F, 0x74, + 0x65, 0x73, 0x74, 0x00, 0x78, 0x65, 0x5F, 0x70, 0x6F, 0x69, 0x6E, 0x74, 0x5F, 0x73, 0x69, 0x7A, 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x32, 0x00, 0xAB, 0xAB, 0xAB, 0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.txt b/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.txt index 0d09ba3be..3288dc41a 100644 --- a/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.txt +++ b/src/xenia/gpu/d3d12/shaders/dxbc/primitive_point_list_gs.txt @@ -9,7 +9,7 @@ // // uint xe_flags; // Offset: 0 Size: 4 [unused] // uint xe_vertex_index_endian; // Offset: 4 Size: 4 [unused] -// uint xe_vertex_base_index; // Offset: 8 Size: 4 [unused] +// int xe_vertex_base_index; // Offset: 8 Size: 4 [unused] // uint xe_pixel_pos_reg; // Offset: 12 Size: 4 [unused] // float3 xe_ndc_scale; // Offset: 16 Size: 12 [unused] // float xe_pixel_half_pixel_offset; // Offset: 28 Size: 4 [unused] diff --git a/src/xenia/gpu/d3d12/shaders/xenos_draw.hlsli b/src/xenia/gpu/d3d12/shaders/xenos_draw.hlsli index e200953d4..014cbe7b4 100644 --- a/src/xenia/gpu/d3d12/shaders/xenos_draw.hlsli +++ b/src/xenia/gpu/d3d12/shaders/xenos_draw.hlsli @@ -5,7 +5,7 @@ cbuffer xe_system_cbuffer : register(b0) { // vec4 0 uint xe_flags; uint xe_vertex_index_endian; - uint xe_vertex_base_index; + int xe_vertex_base_index; uint xe_pixel_pos_reg; // vec4 1 float3 xe_ndc_scale; diff --git a/src/xenia/gpu/dxbc_shader_translator.cc b/src/xenia/gpu/dxbc_shader_translator.cc index 877aade52..0a8a24df8 100644 --- a/src/xenia/gpu/dxbc_shader_translator.cc +++ b/src/xenia/gpu/dxbc_shader_translator.cc @@ -412,7 +412,7 @@ void DxbcShaderTranslator::StartVertexShader_LoadVertexIndex() { // Convert to float and replicate the swapped value in the destination // register (what should be in YZW is unknown, but just to make it a bit // cleaner). - shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_ITOF) | + shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_UTOF) | ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(5)); shader_code_.push_back( EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b1111, 1)); @@ -3026,7 +3026,7 @@ const DxbcShaderTranslator::SystemConstantRdef DxbcShaderTranslator:: // vec4 0 {"xe_flags", RdefTypeIndex::kUint, 0, 4}, {"xe_vertex_index_endian", RdefTypeIndex::kUint, 4, 4}, - {"xe_vertex_base_index", RdefTypeIndex::kUint, 8, 4}, + {"xe_vertex_base_index", RdefTypeIndex::kInt, 8, 4}, {"xe_pixel_pos_reg", RdefTypeIndex::kUint, 12, 4}, // vec4 1 {"xe_ndc_scale", RdefTypeIndex::kFloat3, 16, 12}, diff --git a/src/xenia/gpu/dxbc_shader_translator.h b/src/xenia/gpu/dxbc_shader_translator.h index b65c32f09..e898ab113 100644 --- a/src/xenia/gpu/dxbc_shader_translator.h +++ b/src/xenia/gpu/dxbc_shader_translator.h @@ -276,7 +276,7 @@ class DxbcShaderTranslator : public ShaderTranslator { // vec4 0 uint32_t flags; uint32_t vertex_index_endian; - uint32_t vertex_base_index; + int32_t vertex_base_index; uint32_t pixel_pos_reg; // vec4 1 diff --git a/src/xenia/gpu/dxbc_shader_translator_fetch.cc b/src/xenia/gpu/dxbc_shader_translator_fetch.cc index 5a1ad79f6..25bf6d9d7 100644 --- a/src/xenia/gpu/dxbc_shader_translator_fetch.cc +++ b/src/xenia/gpu/dxbc_shader_translator_fetch.cc @@ -330,18 +330,48 @@ void DxbcShaderTranslator::ProcessVertexFetchInstruction( UpdateInstructionPredication(instr.is_predicated, instr.predicate_condition, true); - // Convert the index to an integer. + // Convert the index to an integer, according to + // http://web.archive.org/web/20100302145413/http://msdn.microsoft.com:80/en-us/library/bb313960.aspx + // (truncating rather than flooring to skip one operation because negatives + // are not valid anyway, and it's safer if -tiny value becomes 0, also doing + // round-to-nearest rather than round_ne so 0.5, 1.5, 2.5, 3.5 is 1, 2, 3, 4, + // which is more valid as an index sequence than 0, 2, 2, 4). DxbcSourceOperand index_operand; LoadDxbcSourceOperand(instr.operands[0], index_operand); - shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_FTOI) | - ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH( - 3 + DxbcSourceOperandLength(index_operand))); - shader_code_.push_back( - EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b0001, 1)); - shader_code_.push_back(system_temp_pv_); - UseDxbcSourceOperand(index_operand, kSwizzleXYZW, 0); - ++stat_.instruction_count; - ++stat_.conversion_instruction_count; + if (instr.attributes.is_index_rounded) { + shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_ADD) | + ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH( + 5 + DxbcSourceOperandLength(index_operand))); + shader_code_.push_back( + EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b0001, 1)); + shader_code_.push_back(system_temp_pv_); + UseDxbcSourceOperand(index_operand, kSwizzleXYZW, 0); + shader_code_.push_back( + EncodeScalarOperand(D3D10_SB_OPERAND_TYPE_IMMEDIATE32, 0)); + shader_code_.push_back(0x3F000000); + ++stat_.instruction_count; + ++stat_.float_instruction_count; + shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_FTOU) | + ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(5)); + shader_code_.push_back( + EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b0001, 1)); + shader_code_.push_back(system_temp_pv_); + shader_code_.push_back( + EncodeVectorSelectOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0, 1)); + shader_code_.push_back(system_temp_pv_); + ++stat_.instruction_count; + ++stat_.conversion_instruction_count; + } else { + shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_FTOU) | + ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH( + 3 + DxbcSourceOperandLength(index_operand))); + shader_code_.push_back( + EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b0001, 1)); + shader_code_.push_back(system_temp_pv_); + UseDxbcSourceOperand(index_operand, kSwizzleXYZW, 0); + ++stat_.instruction_count; + ++stat_.conversion_instruction_count; + } UnloadDxbcSourceOperand(index_operand); // TODO(Triang3l): Index clamping maybe. @@ -375,7 +405,7 @@ void DxbcShaderTranslator::ProcessVertexFetchInstruction( ++stat_.uint_instruction_count; // Calculate the address of the vertex. - shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_IMAD) | + shader_code_.push_back(ENCODE_D3D10_SB_OPCODE_TYPE(D3D10_SB_OPCODE_UMAD) | ENCODE_D3D10_SB_TOKENIZED_INSTRUCTION_LENGTH(9)); shader_code_.push_back( EncodeVectorMaskedOperand(D3D10_SB_OPERAND_TYPE_TEMP, 0b0001, 1)); @@ -390,7 +420,7 @@ void DxbcShaderTranslator::ProcessVertexFetchInstruction( EncodeVectorSelectOperand(D3D10_SB_OPERAND_TYPE_TEMP, 1, 1)); shader_code_.push_back(system_temp_pv_); ++stat_.instruction_count; - ++stat_.int_instruction_count; + ++stat_.uint_instruction_count; // Add the element offset. if (instr.attributes.offset != 0) {